Contextual deconvolution network for semantic segmentation. (May 2020)

• Record Type: Journal Article
• Authors: Fu, Jun; Liu, Jing; Li, Yong; Bao, Yongjun; Yan, Weipeng; Fang, Zhiwei; Lu, Hanqing
• Is Part Of: Pattern recognition. Volume 101(2020:May)
• Publication Date: 2020-05
• Subjects: Semantic segmentation -- Deconvolution network -- Channel contextual module -- Spatial contextual module; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2019.107152
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A Contextual Deconvolution Network (CDN) is proposed to enhance the representation power of the decoder network. A channel contextual module is proposed to model the channel interdependencies of features, which imposes a global semantic control. A spatial contextual module is introduced to model the spatial interdependencies, making the features more expressive on some local regions. The proposed method achieves competitive performance on PASCAL VOC 2012, ADE20K, PASCAL-Context and Cityscapes dataset, and new state-of-the-art performance on PASCAL-Context dataset. Abstract: In this paper, we propose a Contextual Deconvolution Network (CDN) and focus on context association in decoder network. Specifically, in upsampling path, we introduce two types of contextual modules to model the interdependencies of features in channel and spatial dimensions respectively. The channel contextual module captures image-level semantic information by aggregating the feature maps across spatial dimensions, and clarifies global ambiguity of features. Meanwhile, the spatial contextual module obtains patch-level semantic context by learning a spatial weight map, and enhance the feature discrimination. We embed the two contextual modules into individual components of the decoder network, thus improving the representation power and gaining more precise segment results. Thorough evaluations are performed on four challenging datasets, i.e., PASCAL VOC 2012, ADE20K, PASCAL-Context and Cityscapes dataset. Our approach achieves competitive performance with state-of-the-art models on PASCAL VOC 2012, ADE20K and Cityscapes dataset, and new state-of-the-art performance on PASCAL-Context dataset.